1: \documentclass[aps,prl,twocolumn,showpacs,superscriptaddress]{revtex4}
2: % Version History & Changes
3: %
4: % V1 - First draft by K. Delaney. Started 09/09/05. Completed 16/12/05.
5: % V2 - Second draft by K. Delaney. Started 01/24/06.
6: % Fewer technical details plus clarification of conclusions.
7: % Completed: 01/24/06.
8: % V3 - Third draft by K.Delaney. Started 02/13/06.
9: % Small changes from David's feedback. Changed FSE graphs.
10: % Completed: 03/09/06
11: % v4 - Fourth draft by K. Delaney. Started 03/14/06.
12: % Changes from Carlo's feedback.
13: % Changes by DMC on 3.16.06
14: % Further feedback from Carlo on 3.20.06 and 3.22.06
15: % Completed: 3.23.06
16: % v5 - Fifth draft by K. Delaney. Started 05/10/06.
17: % Changes from referees' reports.
18: % v6 - Changes by DMC 5/6-7/06
19: % v7 - K. Delaney 08/06/06. Changes from second round referees' reports.
20: % v8 - DMC 10/12/06 A few changes for next round
21:
22: \usepackage{graphics}
23: \usepackage{amsmath}
24: \usepackage{epsfig}
25:
26: %Shortcut macros
27: \newcommand{\vect}[1]{{\bf {#1}}}
28:
29: \begin{document}
30: %Fill out the title section
31: %\title{Quantum Monte Carlo Study of High-Pressure Hydrogen Fluid and the Plasma Phase Transition}
32: \title{Quantum Monte Carlo Simulation of the High-Pressure Molecular-Atomic Crossover in Fluid Hydrogen}
33:
34: %Author information
35: \author{Kris~T.~Delaney}
36: \affiliation{Department~of~Physics, University~of~Illinois~at~Urbana-Champaign,
37: Urbana, IL 61801, USA}
38: \author{Carlo~Pierleoni}
39: \affiliation{INFM-SOFT and Dipartimento~di~Fisica, Universit\`a~dell'Aquila, I-67010 L'Aquila, Italy}
40: \author{D.~M.~Ceperley}
41: \affiliation{Department~of~Physics, University~of~Illinois~at~Urbana-Champaign,
42: Urbana, IL 61801, USA}
43:
44: \date{\today}
45:
46: \begin{abstract}
47: A first-order liquid-liquid phase transition in high-pressure hydrogen between
48: molecular and atomic fluid
49: phases has been predicted in computer simulations using {\it ab
50: initio} molecular dynamics approaches. However, experiments
51: indicate that molecular dissociation may occur through a continuous crossover rather
52: than a first-order transition. Here we study the nature of molecular
53: dissociation in fluid hydrogen using an alternative simulation
54: technique in which electronic correlation is computed within
55: quantum Monte Carlo, the so-called Coupled Electron Ion Monte
56: Carlo (CEIMC) method. We find no evidence for a first-order
57: liquid-liquid phase transition.
58: \end{abstract}
59:
60: %PACS numbers with commented descriptions
61: \pacs{
62: %61.20.-p% Structure of liquids
63: 61.20.Ja, % Computer simulation of liquid structure
64: 62.50.+p, % High-pressure and shock wave effects in solids and liquids
65: 64.70.Ja, % Liquid-liquid transitions
66: 77.84.Bw% Elements, oxides, nitrides, borides, carbides, chalcogenides, etc.
67: }
68:
69: \maketitle
70:
71: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
72: %%% INTRODUCTION
73:
74: The behavior of hydrogen under a range of thermodynamic conditions
75: is an important problem in theoretical physics, as well as in
76: planetary science and high-pressure physics. Hydrogen exhibits a
77: rich variety of properties including several molecular and
78: non-molecular phases,\cite{HReview} a fluid metal-insulator
79: transition,\cite{shockwave} and possible fluid\cite{Hquantumfluid}
80: and superconducting\cite{Hsuperconduct} phases at low temperature.
81: There exists much uncertainty about the equilibrium properties of
82: high-pressure phases, including the structure and relative
83: stabilities of a number of solid phases,\cite{HSolidPhases,HSolidPhases2} the
84: specific shape of the molecular-solid melting curve
85: and the reason for the density
86: maximum,\cite{Hquantumfluid} and the conditions of molecular
87: dissociation in the fluid.
88:
89: Shock-wave experiments \cite{shockwave} have hinted that the
90: metalization of hydrogen in the liquid state occurs in a
91: continuous manner at temperatures above $3000\mathrm K$. However,
92: recent computer simulations\cite{ScandoloPNAS03,Hquantumfluid}
93: using Car-Parrinello molecular dynamics (CPMD) with density
94: functional theory within the local density approximation (DFT-LDA)
95: and generalized gradient approximation (DFT-GGA) predict that the
96: molecular dissociation in the fluid happens through a first-order phase
97: transition, with discontinuous metalization, at temperatures lower
98: than $3000\mathrm{K}$. These results could be reconciled by having
99: the critical point for the transition at $T < 3000\mathrm{K}$.
100: In addition, other simulations of hydrogen fluid phases, which
101: do not directly address the nature of the molecular dissociation, have been
102: reported \cite{HFluidSimsNOPPT}.
103:
104: In this Letter, we present the results of a study of this
105: transition employing a new quantum Monte Carlo (QMC) method,
106: coupled electron-ion Monte Carlo (CEIMC).\cite{ceimc1,ceimc2} The
107: first QMC approaches for studying high pressure hydrogen were limited to
108: studying solid phases at $T=0\mathrm{K}$: variational Monte Carlo (VMC) and diffusion
109: quantum Monte Carlo (DMC)\cite{T0HHdiss,RPAJastrow}. Finite-temperature
110: restricted path integral Monte Carlo (RPIMC) was used
111: to investigate the possibility of a hydrogen liquid-liquid phase transition
112: \cite{pimcppt}, finding a continuous molecular dissociation at
113: $T \simeq 10000\mathrm{K}$ in agreement with experiments.\cite{shockwave}
114: However, RPIMC becomes inefficient and inaccurate at lower
115: temperatures.
116:
117: The CEIMC approach, in contrast with earlier QMC methods, involves
118: simultaneous evolution of separate but coupled Monte Carlo
119: simulations for the electronic and ionic subsystems, which may
120: then be treated on different levels of approximation. A large gain
121: in computational efficiency can be made for lower temperatures by
122: employing the Born-Oppenheimer approximation (BOA) and requiring
123: that the electronic subsystem remain in its ground state
124: ($T=0\mathrm{K}$). The ionic system is treated at finite
125: temperature either as a set of classical particles, or quantum
126: mechanically using the imaginary-time path integral formalism.
127: This decoupling is good provided the thermal excitation of
128: electrons is a small effect, a fact that holds if $T\ll T_F$,
129: where $T_F$ is the Fermi temperature. For the densities of
130: interest, those at which molecules dissociate in the fluid, $T_F >
131: 200,000\mathrm{K}$. The quality of the BOA applied to high-pressure
132: hydrogen fluid has been tested, and found to be good, by comparing
133: CEIMC to accurate RPIMC calculations, which contain no such
134: approximation, at temperatures of 5,000 and
135: 10,000K.\cite{CarloH2Melt} We therefore anticipate that the BOA
136: will be even more accurate at lower temperatures.
137:
138: For classical nuclei in the canonical ensemble, one can generate
139: the configuration-space probability distribution by proposing a
140: move from configuration $S$\cite{defineS} to configuration
141: $S^\prime$ with uniform probability and using the Metropolis
142: acceptance probability
143: \begin{equation}
144: A=\mathrm{min}\left[1,\mathrm{exp}\left(-\beta\Delta\left(S,S^\prime\right)\right)\right],
145: \end{equation}
146: where $\beta$ is the inverse temperature,
147: $\left(k_BT\right)^{-1}$, and $\Delta\left(S,S^\prime\right)$ the
148: difference in BO energies of nuclear configurations $S$ and
149: $S^\prime$, computed with a $T=0\mathrm{K}$ QMC method. The bias resulting
150: from the statistical noise of $\Delta$ is reduced with correlated
151: sampling\cite{ceimc1,ceimc2}, and eliminated by using the penalty
152: method \cite{penalty}.
153:
154: For electronic sampling, we use either VMC or reptation quantum
155: Monte Carlo (RQMC). RQMC \cite{rqmc} is a projector method that
156: does not suffer from the mixed-estimator bias of DMC and allows
157: efficient calculations of energy differences. To handle electron
158: antisymmetry, we use the fixed-phase method \cite{fixedphase}
159: which allows the modulus of the many-electron wavefunction to be
160: fully optimized while the phase is held fixed. This gives an
161: estimate of the ground state energy much lower than that of the
162: variational estimate (by typically 2500K/atom) but above that of
163: the exact ground state energy. Analysis of related electron
164: systems suggest that the error of the fixed-phase energy will be
165: between 10\% and 30\% of the VMC error,\cite{kwon} i.\,e.,
166: between 250K/atom and 750K/atom. We expect relative errors,
167: between different proton configurations, to be even
168: less.\cite{C184} We also employ twist-averaged boundary conditions
169: (TABC)\cite{TABC} in the electronic calculation to greatly
170: reduce finite-size effects, a problem that has caused substantial
171: errors in earlier simulations of fluid hydrogen.\cite{Hohl}
172: Well-converged energies and correlation functions are achieved
173: using, depending on the density, between 108 and 500 twist angles.
174:
175: In the CEIMC approach, one is free to choose any trial
176: wavefunction for the electronic ground-state. An important
177: consideration is the computer time required per step, since the wavefunction
178: must be calculated for many nuclear configurations. When studying dense
179: liquid hydrogen, we require a general wavefunction that is equally
180: accurate for both molecular and non-molecular configurations.
181: Therefore, we employ a fast band-structure calculation with an
182: effective one-electron potential designed to produce accurate
183: single-particle orbitals for a Slater-Jastrow type wavefunction.
184: Within the fixed-phase approach, it is only the orbitals that
185: affect the systematic bias. One such set of orbitals would be
186: those computed within Kohn-Sham theory. However, full
187: self-consistency for each proton displacement would be too
188: expensive for generating a large number of nuclear
189: configuration-space samples. Consequently, we use a bare
190: electron-proton potential for the effective single-particle
191: potential. Further screening and correlation effects are
192: introduced with a Slater-Jastrow wavefunction; the Jastrow factor
193: is from the RPA pseudopotential, including the one-body (electron-proton)
194: term.\cite{RPAJastrow} Such an approach is surprisingly good, as
195: demonstrated by Figure \ref{fig:wfntest} which shows VMC and RQMC total
196: energies for five different frozen nuclear configurations. This
197: trial function is comparable in quality to one using Kohn-Sham LDA
198: orbitals in the Slater determinant for all configurations and densities tested,
199: while being faster to generate. It is also more transferable than both
200: localized molecular orbitals\cite{ceimc1} (which cannot be used for
201: non-molecular systems) and analytic backflow\cite{backflow} (which is highly accurate
202: only for high-density metallic systems).
203: \begin{figure}
204: \resizebox{\columnwidth}{!}{\rotatebox{-90}{\includegraphics{images/wfnall.ps}}}
205: \caption{VMC and RQMC total energies for five different crystal configurations at two different
206: densities (left graphs $V=9.42 \mathrm{a.u./atom}$ or $r_s=1.31$; right graphs $V=33.51 \mathrm{a.u./atom}$ or $r_s=2.0$)
207: using a number of different Slater-Jastrow wavefunctions. Configuration 1 is molecular,
208: 2--5 are non-molecular. Owing to the variational principle, a lower energy implies a
209: better wavefunction. DFT-LDA and bare electron-proton bands (see text) provide the
210: best and most transferable orbitals for the Slater determinant. Gaussian for configuration
211: 1 refers to localized molecular orbitals.}
212: \label{fig:wfntest}
213: \end{figure}
214:
215: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
216: %%% Results
217: We investigate the atomic-molecular transition in liquid hydrogen
218: at fixed density ($1.35 \leq r_s \leq 1.55$ where $4 \pi
219: r_s^3/3=1/n_e$) and temperature ($T=2000\mathrm{K}$ and $1500\mathrm{K}$).
220: The pressure, estimated using the virial theorem, an approach that is accurate
221: with RQMC due to the lack of a mixed-estimator bias, lies between
222: 135GPa and 290GPa for this range of densities.
223: Using a classical Monte Carlo simulation with
224: an empirical potential,\cite{ceimc2} the system is prepared
225: either in a purely molecular or a purely atomic fluid initial
226: configuration. During the subsequent simulation the system evolves
227: to its equilibrium state, subject to overcoming any free-energy
228: barriers during the simulation. A typical simulation has 14,000
229: ionic moves with a step size of 0.006--0.016Bohr. We collect
230: statistics along the sequence of ionic states, particularly the
231: proton-proton correlation functions which give insight into the
232: state of the liquid through a distinctive peak at
233: $r_{pp} \sim 1.4 \mathrm{Bohr}$ when molecules are present. Figure
234: \ref{fig:t2kgrs} shows a set of proton-proton correlation
235: functions for simulations at several densities prepared with
236: either a molecular or atomic fluid as the initial state.
237:
238: \begin{figure}
239: \resizebox{\columnwidth}{!}{{\rotatebox{-90}{\includegraphics{images/T2000KAllGrs.ps}}}}
240: \caption{Proton-proton correlation functions for CEIMC simulations in the canonical ensemble with 32 atoms at $T=2000\mathrm{K}$. The BO energies are computed with VMC (left pane) or RQMC (right pane) and the wavefunction is a Slater-Jastrow type with bare electron-proton bands (see text). Grey lines are simulations initially prepared with a molecular fluid and black with an atomic fluid. Hysteresis is evident. $P_\mathrm{at}$ and $P_\mathrm{m}$ are the computed pressures for simulations prepared with an atomic and a molecular fluid respectively. All correlation functions are plotted to the same scale.}
241: \label{fig:t2kgrs}
242: \end{figure}
243:
244: For a quantitative analysis, a method for estimating the average
245: number of molecules from the pair-correlation function at each phase point
246: is required. A rough estimate would involve integrating the pair-correlation
247: function up to the first minimum, but such an
248: approach does not take into account the ``baseline'' caused by nearby molecules and unbound atoms; see Fig. \ref{fig:t2kgrs}. To remove the baseline, we fit each of
249: the pair-correlation functions to the functional form:
250: \begin{equation}
251: g\left(r\right) = \lambda g_\mathrm{m}\left(r;\left\{\alpha\right\}\right) + \left(1-\lambda\right)g_\mathrm{at}\left(r;\left\{\gamma\right\}\right),
252: \label{eq:mop}
253: \end{equation}
254: where $\left\{\alpha,\gamma\right\}$ are fitting parameters and
255: $g_\mathrm{m}$ is a Gaussian centered on 1.3--1.5Bohr.
256: $g_\mathrm{at}$ is a fit to the next-nearest neighbor
257: distribution.
258: We find that the quality of the fit is insensitive to the choice of
259: $g_\mathrm{at}$; the molecular fraction varies by no more than
260: $3\%$ for different choices. The molecular order parameter,
261: $\lambda$, the fraction of protons that are bound into
262: a molecule, is plotted against pressure in Fig.
263: \ref{fig:orderparam} for $T=2000\mathrm{K}$.
264:
265: We find that the CEIMC simulations using VMC energy differences
266: (left pane Fig.~\ref{fig:t2kgrs}, and Fig.~\ref{fig:orderparam}) yield an
267: irreversible transition with hysteresis. This implies a
268: free-energy barrier that is difficult to overcome during the
269: simulations, hence a metastable state is obtained. The large width of the
270: hysteresis loop is indicative of a transition that is weakly
271: first-order. We therefore expect that $T=2000\mathrm{K}$ is close to the VMC
272: critical point.
273:
274: In contrast, the much more accurate RQMC simulations (right pane
275: Fig.~\ref{fig:t2kgrs}, and Fig.~\ref{fig:orderparam}) yield a reversible and
276: continuous molecular dissociation upon increasing pressure. Both
277: molecular and atomic states are mechanically unstable in the range
278: of densities simulated, and the system quickly evolves, with no
279: discernible free-energy barrier, to a stable part-molecular,
280: part-atomic state. This behavior implies the lack of an underlying
281: first-order transition, a behavior which would not change as the
282: thermodynamic limit is approached. The likely reasons for the
283: change in character of the crossover is that the BO energy
284: surface of the assumed Slater-Jastrow trial function is inaccurate
285: under conditions at which molecules are short-lived transient entities.
286: As discussed above, these deficiencies are corrected by the
287: imaginary-time projection of RQMC, if they are due to errors in the
288: modulus of the trial wavefunction and not its phase.
289:
290: \begin{figure}
291: \resizebox{0.85\columnwidth}{!}{\rotatebox{-90}{\includegraphics{images/T2000KHysteresis.ps}}}
292: \caption{Molecular order parameter ($\lambda$ in Eq.\ \ref{eq:mop}) evaluated for simulation results from Fig.\ \ref{fig:t2kgrs}. Two curves are obtained for each method from different initial conditions (see text). VMC (dashed lines) simulations indicate an irreversible, weakly first-order phase transition for molecular dissociation in the fluid with increasing/decreasing pressure. This picture is compatible with conclusions from CPMD DFT-LDA. Accurate RQMC (solid lines) simulations find a reversible crossover.}
293: \label{fig:orderparam}
294: \end{figure}
295:
296: We have assessed the finite-size errors in the cell volume by
297: repeating a selection of simulations with 54 atoms. The
298: proton-proton correlation functions in Figure \ref{fig:fse} show
299: the comparison between 32- and 54-atom simulations for VMC and
300: RQMC. For VMC finite-size errors are clearly large, consistent
301: with the presence of a free-energy barrier, while for RQMC the errors are
302: small, indicative of a continuous crossover between molecular and
303: atomic states. %The latter is the expected
304: %behavior for a fluid, when lacking a first-order phase transition, due to
305: %lack of long-range order.
306: %positional and, in a molecular state, orientational disorder.
307: \begin{figure}
308: \resizebox{0.65\columnwidth}{!}{{\includegraphics{images/T2000KVMCFSE.ps}}}
309: \resizebox{0.65\columnwidth}{!}{{\includegraphics{images/T2000KRQMCFSE.ps}}}
310: \caption{Demonstration of finite-size errors in VMC pair-correlation functions (upper pane). Solid lines are 32-atom simulations, dashed are 54 atoms. Grey lines are simulations started from a molecular fluid and black are from an atomic fluid. Quoted pressures are for simulations started from a molecular fluid. RQMC (lower pane) simulations show small finite-size errors.}
311: \label{fig:fse}
312: \end{figure}
313:
314:
315: We have tested the nature of molecular dissociation at a lower
316: temperature: $T=1500\mathrm{K}$. Figure \ref{fig:t1500grs} shows
317: pair-correlation functions. Hysteresis is again evident for VMC
318: simulations, although over a shorter pressure-range, indicating a
319: larger free-energy barrier. Again, RQMC simulations do not show
320: a first-order transition, mirroring the behavior at
321: $T=2000\mathrm{K}$.
322:
323: %RQMC simulations again demonstrate a
324: %reversible and continuous transition, mirroring the behavior at
325: %$T=2000\mathrm{K}$.
326:
327: \begin{figure}
328: \resizebox{\columnwidth}{!}{\rotatebox{-90}{\includegraphics{images/T1500KGrs.ps}}}
329: \caption{Proton-proton correlation functions for simulations at $T=1500\mathrm{K}$. Black lines are simulations prepared with an atomic fluid, grey are prepared with a molecular fluid. The simulations demonstrate an irreversible transition with hysteresis for VMC. Left pane: VMC; Right pane: RQMC.}
330: \label{fig:t1500grs}
331: \end{figure}
332:
333: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
334: %%% CONCLUSIONS
335:
336: In conclusion, this study demonstrates that the CEIMC approach
337: with a fast and accurate trial wavefunction can be used to
338: determine the nature of the pressure-driven molecular to atomic
339: crossover in hydrogen for temperatures on the order of 1000K. We
340: find that the improvements in interatomic interactions obtained
341: when using RQMC to simulate the electronic subsystem lead to no
342: first-order phase transition, a result of fundamentally different
343: nature to that predicted when using VMC. Our most accurate simulations,
344: those employing RQMC, provide no evidence for a first-order atomic-molecular
345: transition in the liquid phase at either $T=1500\mathrm{K}$
346: or $T=2000\mathrm{K}$, showing instead a continuous crossover;
347: this is in contrast to the first-order transition found
348: with CPMD. The remaining uncontrolled approximations in our
349: simulations are the finite-size error (demonstrated to be small
350: for RQMC), the fixed-phase approximation in RQMC, the adiabatic
351: approximation for decoupling nuclear and electronic time-scales
352: (tested in ref. \cite{CarloH2Melt}) and the use of classical
353: statistics for simulating nuclei. Path integral calculations are
354: in progress for studying the effect of nuclear quantum statistics
355: on dense liquid hydrogen. We anticipate that nuclear quantum
356: effects will destabilize molecules at a lower pressure leaving the
357: crossover character unchanged.
358:
359: \begin{acknowledgments}
360: This material is based upon work supported by the NSF awards DMR 04-04853
361: and DMR 03-25939 ITR, and by MIUR-COFIN2003. Calculations were
362: performed at the NCSA at the University of Illinois at
363: Urbana-Champaign and at CINECA (Italy). We thank S. Chiesa and D.
364: Quigley for fruitful discussions.
365: \end{acknowledgments}
366:
367: \begin{thebibliography}{10}
368: \bibitem{HReview}H.-K.~Mao and R.J.~Hemley, Rev. Mod. Phys. {\bf 66}, 671 (1994).
369: \bibitem{shockwave}S.T.~Weir {\it et al.}, Phys. Rev. Lett. {\bf 76}, 1860 (1996).
370: \bibitem{Hquantumfluid}S.A.~Bonev {\it et al.}, Nature {\bf 431}, 669 (2004).
371: \bibitem{Hsuperconduct}E.~Babaev {\it et al.}, Nature {\bf 431}, 666 (2004).
372: \bibitem{HSolidPhases} V.~Natoli, R.M.~Martin and D.~Ceperley, Phys. Rev. Lett. {\bf 74}, 1601 (1995); K.A.~Johnson and N.W.~Ashcroft, Nature {\bf 403}, 632 (2000).
373: \bibitem{HSolidPhases2} S.~Biermann {\it et al.}, Solid State Comm. {\bf 108}, 337 (1998); Jorge~Kohanoff {\it et al.}, Phys. Rev. Lett. {\bf 78}, 2783 (1997); Jorge~Kohanoff {\it et al.}, Phys. Rev. Lett. {\bf 83}, 4097 (1999).
374: \bibitem{ScandoloPNAS03}S.~Scandolo, PNAS {\bf 100}, 3051 (2003).
375: \bibitem{HFluidSimsNOPPT}Ali~Alavi {\it et al.}, Phys. Rev. Lett. {\bf 73}, 2599 (1994); Jorge~Kohanoff {\it et al.}, Phys. Rev. E {\bf 54}, 768 (1996); Stefan~Nagel {\it et al.}, Phys. Rev. E {\bf 57}, 5572 (1998).
376: \bibitem{ceimc1}M.~Dewing and D.M.~Ceperley, in {\it Recent Advances in quantum Monte Carlo Methods, II}, edited by S.~Rothstein (World Scientific, Singapore, 2003).
377: \bibitem{ceimc2}D.M.~Ceperley, M.~Dewing and C.~Pierleoni, {\it Bridging Time Scales}, edited by P.~Niebala {\it et al.}, Lecture Notes in Physics Vol. 605 (Springer-Verlag, Berlin, 2003); C.~Pierleoni and D.M.~Ceperley, to appear in ``Lecture Notes in Physics'' (2006), arXiv:physics/0510254.
378: \bibitem{T0HHdiss}D.M.~Ceperley and B.J.~Alder, Phys. Rev. B {\bf 36}, 2092 (1987).
379: \bibitem{RPAJastrow}D.M.~Ceperley and B.J.~Alder, Physica {\bf 108B}, 875 (1981).
380: \bibitem{pimcppt}B.~Militzer and D.M.~Ceperley, Phys. Rev. E {\bf 63}, 066404 (2001).
381: W.R.~Magro {\it et al.}, Phys. Rev. Lett. {\bf 76}, 1240 (1996).
382: \bibitem{CarloH2Melt}C.~Pierleoni, D.M.~Ceperley and M.~Holzmann, Phys. Rev. Lett. {\bf 93}, 146402 (2004).
383: \bibitem{defineS} $S$ is the collection of all
384: nuclear coordinates in the simulation, $S=\left\{{\bf R}_i\right\}$,
385: where ${\bf R}_i$ is the position of the
386: $i^\mathrm{th}$ nucleus.
387: \bibitem{penalty}D.M.~Ceperley and M.~Dewing, J.~Chem.~Phys. {\bf 110}, 9812 (1999).
388: \bibitem{rqmc}S.~Baroni and S.~Moroni, Phys.~Rev.~Lett. {\bf 82}, 4745 (1999).
389: \bibitem{fixedphase}G.~Ortiz, D.M.~Ceperley and R.M.~Martin, Phys. Rev. Lett. {\bf 71}, 2777 (1993).
390: \bibitem{TABC}C.~Lin, F.H.~Zong and D.M.~Ceperley, Phys. Rev. E {\bf 64}, 016702 (2001).
391: \bibitem{kwon} Y.~Kwon, D.M.~Ceperley and R.M.~Martin, Phys. Rev. B {\bf 58}, 6800 (1998).
392: \bibitem{C184} C.~Pierleoni and D.M.~Ceperley, Chem. Phys. Chem. {\bf 6}, 1872 (2005).
393: \bibitem{Hohl}D.~Hohl {\it et al.}, Phys. Rev. Lett. {\bf 71}, 541
394: (1993).
395: \bibitem{backflow}M.~Holzmann {\it et al.}, Phys. Rev. E {\bf 68}, 046707 (2003).
396: \end{thebibliography}
397: \end{document}
398: